以影像分割及邊界描述為基礎的影像壓縮技術

Image Compression by Segmentation and Boundary Description

黃俊德(Jiun-De Huang)

指導教授 ： 丁建均

國立臺灣大學/電機資訊學院/電信工程學研究所/碩士(2008年)

https://doi.org/10.6342/NTU.2008.01461

摘要

在現有的影像壓縮技術上，如JPEG，皆是對整張圖做相同的處理，而不會對於不同的影像內容而有所調整，所以壓縮率有其極限。而在新一代的影像壓縮技術中，是以影像分割為基礎，將影像盡可能的切割成數個特性或色彩值近似的區塊，每個區塊分別有不同的形狀與色彩值。由於同區塊中的色彩值通常會有高度相關，所以理論上可以產生更高的壓縮率。
對於影像分割，大致上是根據像素值的兩種性質：不連續性與相似性。為了找到不連續的像素值，我們將會介紹影像的邊緣偵測的基本技術並且提出一種結合了傳統的微分法與希爾伯轉換法的可適性方法，叫做短時響應的希爾伯轉換。我們也將會介紹許多其他的方法來做影像切割。我們主要的目的是找出適合的分割結果讓壓縮可以更有效率。
做好影像切割之後，我們會介紹JPEG標準中使用到的壓縮演算法，並應用在我們提出的方法中。為了有效率的記錄每個影像區塊的輪廓，我們先介紹一些常用的邊界描述子並提出兩種改進過後的邊界描述子。為了壓縮影像區塊的色彩值，我們將會討論如何將一個不規則形狀的影像區塊轉換到頻域，接著我們便可以對轉換後的頻率係數做量化及編碼來降低資料量。最後我們將結果與使用JPEG標準壓縮的圖片做比較，證實在可接受的失真範圍下，壓縮率的確可以增加很多。

關鍵字

邊緣偵測 ； 影像切割 ； 影像壓縮 ； 傅利葉描述子 ； 不規則型狀的離散餘弦轉換

並列摘要

The present technique of image compression, like the JPEG standard, makes the same process to whole image and does not adjust the parameters based on the local characteristics of the image. Therefore, it has a limit to its compression ratio. However, a new compression technique called segmentation-based image compression has been developed. It segments an image to several regions with similar characteristic or color. Because each image segment has different shapes and color values, we compress these regions individually. Due to the high correlation of the color values in an image segment, we could achieve higher compression ratio in theory.
The technique of image segmentation is based on two properties of color values: discontinuity and similarity. To find the discontinuity of the color values, we will intro-duce the image edge detection technique and propose an adaptive method called the short response Hilbert transform (SRHLT) which combines the traditional differential method and the Hilbert transform method. We will also discuss many other ways to segment an image. The main object is to find a suitable segmented result that can be compressed efficiency.
After Segmentation, we will introduce the basic image compression algorithms in JPEG standard and apply them in our proposed methods. To record the boundary of an image segment efficiently, we will introduce some popular boundary descriptors and propose two improved boundary descriptors. To compress the color values of an image segment, we will discuss how to transform an arbitrary-shape image segment to fre-quency domain. Then we can quantize and encode the frequency coefficients to de-crease the information quantity. Finally, we will compare the result with that of JPEG standard and prove that the compression ratio could be increase a lot under acceptable distortion.

並列關鍵字

Edge detection ； Image segmenation ； Image compression ； Fourier description ； Arbitrary-shape DCT

參考文獻

A. Digital Image Processing

[3] T. Acharya, A. K. Ray, Image Processing Principles and Applications, John Wiley & Sons, New Jersey.

[4] W. K. Pratt, Digital Image Processing Third Edition, John Wiley & Sons, Man-hattan, 2002.

B. Image Compression

[7] G. K. Wallace, “The JPEG Still Picture Compression Standard,” Communications of the ACM, vol. 34, issue 4, pp. 30-44, 1991.